Never emit lifetime substitutions

[RyanGlScott]

Currently, mir-json is rather inconsistent about when it erases Lifetime in compiled code or not. Here is an example to illustrate the problem:

pub struct S<'a> {
    pub x: &'a u32,
}

pub fn f<'a>(y: &'a u32) -> S<'a> {
    S { x: y }
}

When the compiled MIR for f is pretty-printed, it looks like this:

fn test/7ab66983::f[0](_1 : &u32) -> test/7ab66983::S[0]<lifetime> { ... }

Note the <lifetime> substitution in the occurrence of S. This is unfortunate, as lifetimes are meant to be erased during compilation, and their presence in the MIR code means that SAW users have to manually instantiate S with lifetime to get what they want. What's more, not all types carry around explicit lifetime substitutions (note that the MIR JSON contains &u32, not &'a u32).

lifetimes are currently emitted here:

mir-json/src/analyz/ty_json.rs

Lines 660 to 663 in ec2cbf6

	// In crucible-mir, all substs entries are considered "types", and there are dummy
	// TyLifetime and TyConst variants to handle non-type entries. We emit something that
	// looks vaguely like an interned type's ID here, and handle it specially in MIR.JSON.
	ty::subst::GenericArgKind::Lifetime(_) => json!("nonty::Lifetime"),

We should strive to avoid emitting lifetime substitutions at all to prevent users from having to care about them.